Background: Congenital long QT syndrome (LQTS) is an uncommon genetic arrhythmia syndrome characterized by a prolonged QT-interval that can lead to syncope, seizures, or sudden cardiac death (SCD). Risk stratification is extremely important to identify patients at highest risk for a LQTS-associated cardiac event (CE), and current scoring systems have limited power. Supervised machine learning (SML), a form of artificial intelligence (AI), may enhance risk stratification in LQTS.
Objective: Utilize AI to build a risk prediction model using known LQTS risk factors and compare SML-guided risk stratification to conventional LQTS risk models.
Methods: 1,309 patients with confirmed LQTS (58% females; mean age at diagnosis 20.1 ± 17.6 years; including 583 with LQT1, 415 with LQT2, and 141 with LQT3) and with the input features of gender, age at diagnosis, QTc, genotype, proband status, and family history of LQTS/SCD, were used to build a risk prediction model. The model was trained on 80% of the cohort (n=1,047) and tested on the remaining 20% (n=262). The model performance was determined by the receiver operating characteristic area under the curve (AUC).
Results: Overall, the mean QTc was 469 ± 40 ms with 201 (15%) patients having a QTc ≥ 500 ms. Over one-third (n=449) were designated as the proband, and 456 (35%) had experienced ≥ 1 LQTS-associated CE. Following training, the model was tested on the 20% holdout dataset and the AI-guided model accurately predicted a possible LQTS-CE with an AUC of 0.82 (95% CI: 0.77 - 0.89), sensitivity 76% (67 - 85), specificity 76% (68 - 84), and accuracy of 76% (70 - 81). This model significantly outperformed previously developed models that used gender, QTc ≥ 500 ms, and LQTS genotypes (AUC 0.68, 95% CI: 0.62 - 0.74, p < 0.0001).
Conclusion: AI-guided risk stratification outperformed current LQTS risk models, and with further testing, AI-LQTS may prove useful in guiding therapy.